Assigning sentiment labels to documents is, at first sight, a standard multi-label classification... more Assigning sentiment labels to documents is, at first sight, a standard multi-label classification task. Many approaches have been used for this task, but the current state-of-the-art solutions use deep neural networks (DNNs). As such, it seems likely that standard machine learning algorithms, such as these, will provide an effective approach. We describe an alternative approach, involving the use of probabilities to construct a weighted lexicon of sentiment terms, then modifying the lexicon and calculating optimal thresholds for each class. We show that this approach outperforms the use of DNNs and other standard algorithms. We believe that DNNs are not a universal panacea and that paying attention to the nature of the data that you are trying to learn from can be more important than trying out ever more powerful general purpose machine learning algorithms.
Many natural language processing (NLP) applications require the computation of similarities betwe... more Many natural language processing (NLP) applications require the computation of similarities between pairs of syntactic or semantic trees. Many researchers have used tree edit distance for this task, but this technique suffers from the drawback that it deals with single node operations only. We have extended the standard tree edit distance algorithm to deal with subtree transformation operations as well as single nodes. The extended algorithm with subtree operations, TED+ST, is more effective and flexible than the standard algorithm, especially for applications that pay attention to relations among nodes (e.g. in linguistic trees, deleting a modifier subtree should be cheaper than the sum of deleting its components individually). We describe the use of TED+ST for checking entailment between two Arabic text snippets. The preliminary results of using TED+ST were encouraging when compared with two string-based approaches and with the standard algorithm.
This paper discusses the design of a planner whose intended application required us to solve the ... more This paper discusses the design of a planner whose intended application required us to solve the so-called 'ramification problem'. The planner was designed for the purpose of planning communicative actions, whose effects are famously unknowable and unobservable by the doer/speaker, and depend on the beliefs of and inferences made by the observer/hearer. Our fully implemented model can achieve goals that do not match action effects, but that are rather entailed by them, which it does by reasoning about how to act: state-space planning is interwoven with theorem proving in such a way that a theorem prover uses the effects of actions as hypotheses. 1
Among the key issues in computational semantics, the notion that meanings should be seen as conte... more Among the key issues in computational semantics, the notion that meanings should be seen as context changing operations and the need to deal with utterances with multiple readings have received a great deal of attention. Approaches to both topics tend to involve inventing new logics. I believe that the move to non-standard logics should be taken only as a very last resort | changing the nature of the representation is a much bigger step than adding new content within an existing representational framework. In the current paper I will show how to use a simple doxastic logic to deal with both these phenomena.
An account of discourse connectives is given which focusses on their entailments. It covers uses ... more An account of discourse connectives is given which focusses on their entailments. It covers uses of such connectives both to express discourse-internal situational relations and to express meta-level discourse relations. Representing the second of these involves explicitly introducing utterance mood into the logical form. The entailments of both situational and discourse uses of connectives are captured in meaning postulates. Those for situational relations concern only entities in the common ground and are presuppositional. Those for discourse relations may also refer, beyond the common ground, to the beliefs and attitudes of speaker and hearer, and they are informative rather than presuppositional. We suggest that in many cases the situational and discourse interpretations of a single relation have extremely similar properties, and that this shows up when the meaning postulates for the relation are elaborated.
There has been a substantial move recently to take the role of inference seriously in computation... more There has been a substantial move recently to take the role of inference seriously in computational semantics. For a long time there was very little difference between for-mal semantics and computational semantics: in both cases, the task was to construct some formal entity which captured ...
Most optimisation techniques for theorem provers for rst-order logic rely on static analysis of t... more Most optimisation techniques for theorem provers for rst-order logic rely on static analysis of the problem statement. For intensional logics, such as static analysis cannot be relied on, since it is impossible to predict what literals may be introduced by the intensional rules. The current paper shows how to use a dynamic (run-time) version of one well-known static optimisation, and considers its relationship to the use of`relevance checking' in Satchmo.
Uploads
Papers by Allan Ramsay